DETECTION SYSTEM FOR UNDER-AGE VOTERS USING CONVOLUTIONAL NEURAL NETWORK (CNN) AND IRIS DATA

Abstract

Underage voting continues to weaken the credibility of elections in Nigeria, even after the introduction of the Bimodal Voter Accreditation System (BVAS), because voter-age verification still depends on declared age, documents, and manual checks that can be falsified or pressured. This paper presents a detection system for underage voters that classifies iris images into two classes, adult (18 years and above) and under 18, using a Convolutional Neural Network (CNN). An integrated methodology comprising experimental research, a machine learning pipeline, and the Object-Oriented Analysis and Design Method (OOADM) was adopted. Iris images drawn from the CASIA, MMU, and Kaggle repositories (10,000 images; 70:10:20 training, validation, and testing split) were segmented, normalized with Daugman’s rubber sheet model, and passed to an AlexNet-based CNN with batch normalization, dropout, and a two-neuron Softmax output, trained for 300 epochs with the Adam optimizer. The model was implemented with Python for the machine learning back end, JavaScript and PHP for the application programming interface (API) at the front end, and MySQL for database management. Performance was evaluated with a confusion matrix and compared with the iris-based age and gender detector of Faluyi et al. The illustrative 300-case evaluation produced 98.33% accuracy, 97.03% precision, 98.00% recall, 97.51% F1-score and a 1.5% false positive rate, against 94.38% accuracy, 96.18% precision, 94.37% recall and a 5.61% false positive rate for the existing system, while processing time fell from 44.23 s to 28.57 s. The results indicate that CNN-based iris analysis can serve as a decision-support layer for voter-age verification, subject to validation in real registration environments.